Skip to main content

Full text of "Superfluidity vs Bose-Einstein condensation in a Bose gas with disorder"

See other formats


Superfluidity vs Bose-Einstein condensation in a Bose gas with disorder 



G.E. Astrakharchik 1 , J. Boronat 2 , J. Casulleras 2 and S. Giorgini 1 

1 Dipartimento di Fisica, Universitd di Trento, and Istituto Nazionale per la Fisica della Materia, 1-38050 Povo, Italy 
2 Departament de Fisica i Enginyeria Nuclear, Campus Nord B4--B5, Universitat Politecnica de Catalunya, E-08034 

Barcelona, Spain 
(February 1, 2008) 

We investigate the phenomenon of Bose-Einstein condensation and superfluidity in a Bose gas 
at zero temperature with disorder. By using the Diffusion Monte-Carlo method we calculate the 
superfluid and the condensate fraction of the system as a function of density and strength of disorder. 
In the regime of weak disorder we find agreement with the analytical results obtained within the 
Bogoliubov model. For strong disorder the system enters an unusual regime where the superfluid 
fraction is smaller than the condensate fraction. 



PACS numbers: 03.75.Fi, 05.30. Fk, 67.40. Db 



The study of disordered Bose systems has attracted 
in the recent past considerable attention both theoreti- 
cally and experimentally. The problem of boson localiza- 
tion, the superfluid-insulator transition and the nature of 
elementary excitations in the presence of disorder have 
been the object of several theoretical investigations |l]] 
and Monte-Carlo numerical simulations , both based 
on Hubbard or equivalent models on a lattice. More re- 
cently, the problem of Bose systems with disorder has 
also been addressed in the continuum. On the one hand, 
the dilute Bose gas with disorder has been studied within 
the Bogoliubov model On the other, Path Integral 

Monte-Carlo (PIMC) techniques have been applied to the 
study of the elementary excitations in liquid 4 He @ and 
the transition temperature of a hard-sphere Bose gas || , 
in the presence of randomly distributed static impurities. 
Disordered Bose systems are produced experimentally in 
liquid 4 He adsorbed in porous media, such as Vycor or 
silica gels (aerogel, xerogel). The suppression of super- 
fluidity and the critical behavior at the phase transition 
have been investigated in these systems in a classic se- 
ries of experiments and the elementary excitations 
of liquid 4 He in Vycor have been recently studied using 
neutron inelastic scattering p0| . Furthermore, the recent 
achievement of Bose-Einstein condensation (BEC) in al- 
kali vapours has sparked an even larger interest in the 
physics of degenerate Bose gases and their macroscopic 
quantum properties, such as long-range order and super- 
fluid behavior (for a review see fll||). 

In this Letter we investigate the effects of disorder on 
BEC and superfluidity in a Bose gas at zero temperature. 
As a model for disorder a uniform random distribution 
of static impurities is assumed. This choice provides us 
with a reasonable model for 4 He adsorbed in porous me- 
dia and might also be relevant for trapped Bose conden- 
sates in the presence of heavy impurities. In addition, the 
qucnchcd-impurity model allows us to derive analytical 
results in the weak-disorder regime and can be imple- 
mented in a quantum Monte Carlo simulation. 

The present work is divided in two parts. In the first 



part, following the analysis of Ref. the properties of 
the system are investigated within the Bogoliubov ap- 
proximation. Results for the effects of disorder on the 
ground-state energy, superfluid density and condensate 
fraction are discussed. In the second part, we resort 
to the Diffusion Monte-Carlo (DMC) technique, which 
solves exactly the many-body Schrodinger equation for 
the ground state of a boson system. By using this tech- 
nique, we verify that the results of the Bogoliubov model 
apply only to dilute systems with weak disorder and we 
investigate the cross-over to the regime of strong disor- 
der, where the suppression of superfluidity and BEC due 
to the random potential is large. In this regime we find 
that the system exhibits the unusual feature of a super- 
fluid component smaller than the condensate component. 

Bogoliubov model. The starting point is the Bogoliubov 
Hamiltonian of a homogeneous dilute Bose gas 

H = E + ^2 t p a p a p , (1) 
p 

written in terms of the quasi-particle annihilation and 
creation operators a p , a p . These operators are related 
to the particle operators a p , a p through the well-known 
canonical transformation a p — u p a p + v p a^_ p , with co- 
efficients — 1 + v 2 = (e p + gno + e p )/2e p and u p v p = 
—gno/2e p . The elementary excitation energies obey the 
usual Bogoliubov spectrum e p = [(ep 1 ) 2 + 2gno€ p ] 1 ^ 2 , with 
e p = p 2 /2m the free particle energy, n the condensate 
density and g = Airh 2 a/m the coupling constant fixed 
by the s-wave scattering length a. The constant term 
E /N = [Anna 3 + 5120F(na 3 ) 3 / 2 /15]?i 2 /(2ma 2 ) is the 
ground-state energy per particle expressed in terms of 
the gas parameter no? , with n = N/V the total particle 
density. This result includes the zero-point motion of the 
elementary excitations. 

Disorder is introduced in the system by adding to 
Ho the perturbation H' = J d 3 r V(r)n(r) produced 

by the external field V(r) — X)i=? P v (\ r ~ r i\) associ- 
ated with the impurities. Here, N- lmp counts the im- 



f 



purities with fixed position and v(r) is the two- 
body particle-impurity potential. For dilute systems 
and small concentrations of impurities the pair poten- 
tial v(r) can be expressed as a pseudo-potential v(r) = 
Simp<5(r). The coupling constant g; mp = 2-Kh 2 b/m is 
fixed by the particle-impurity s-wave scattering length 
b and by the reduced mass of the pair, which coin- 
cides with the particle mass m if the impurity is in- 
finitely massive. Assuming a uniform random distri- 



bution of impurities with density ni, 



N imp /V and 



gaussian correlated disorder, we obtain that the statis- 
tical properties of disorder are described by the aver- 



age value (V Q ) = 1/V J d 3 r (V(r)) = g 



, and by 



lmp'^imp 

the correlation function C(s) = 1/V J d 3 r (V(r)V(r + 
s)), whose Fourier transform is given by (V P V- P ) = 
1/V J d 3 s e~ l P s / h C(s) = gf mp n imp /V. The notation (..) 
stands here for average over disorder configurations. The 
model is described by three parameters: i) the gas param- 
eter na 3 , ii) the concentration of impurities \ — N[ mp /N, 
and iii) the ratio of scattering amplitudes b/a. The first 
parameter is related to the strength of interactions, the 
other two to the strength of disorder. Within the Bo- 
goliubov model all relevant properties of the system de- 
pend on disorder through the combination R — x (b/a) 2 , 
which gives a measure of the strength of disorder. 

The perturbation term H' can be written in momen- 
tum space as H' = NVq + J2 p V-pPpi where p p is 
the density fluctuation operator. Within the Bogoli- 
ubov approximation we write p p ~ \/No (a p + aL p ) = 

V No (u p + v p )(a p + cr_), where Nq is the number 
of atoms in the condensate. The total Hamiltonian 
H = Hq + H' is given by a combination of linear and 
quadratic terms in the quasi-particle operators a p , ap 
and can be diagonalized by means of the operator shift 
§ a p = P P - y/No~V p (u p + Vp)le p . One finds 



H = E 



(2) 



To lowest order, the elementary excitation energies are 
not affected by the random field, whereas the ground- 
state energy is given by E = E Q + N[g imp n imp - 
0? mp rcimp(lAOEp 2m /(P 2 + 4m 5«o)]- The term pro- 
portional to gf mp is ultraviolet divergent, but the diffi- 
culty is overcome if one takes into account the second or- 
der correction to the particle-impurity coupling constant 
.gimp -» Simp + Simpl 1 /^) Dp 2ra/p 2 . The final result for 
the ground-state energy per particle in units of h 2 /2ma 2 
reads 



E 
N 



N 



+ (na 3 ) 3 / 2 



15 



16?r 3/2 R 



(3) 



where Emf/N = 4n(na 3 )[l + x (b/a,)] is the mean-field 
contribution. Notice that the model of ^-correlated dis- 
order of Refs. does not allow the calculation of the 
ground-state energy, since the renormalization of gi mp is 
a crucial step. 



The depletion of the condensate and the non-superfluid 
component of the gas can be obtained from the Hamilto- 
nian (0) by calculating, respectively, the momentum dis- 
tribution and the long-wavelength behavior of the static 
transverse current-current response function |^,^|. For 
the condensate fraction one finds 



Np 8 
N 30? 



(na 3)l/2_ ^(^3)1/2^ (4) 



in which the first term gives the quantum depletion due 
to interaction and the second term accounts for the ef- 
fect of disorder. Differently from No/N, the superfluid 
fraction is equal to unity in the absence of disorder and 
one has 



^ = l_!^ (na 3)l/ 2i? 

p 3 2 l ' 



(5) 



As it has been anticipated, both the result for the energy 
beyond mean field (||) and results (^) and (||) depend on 
disorder through the scaling parameter R = x (b/a) 2 . 
Another interesting consequence of the above results is 
that, due to the coefficient 4/3 in (|J), disorder is more 
efficient in depleting the superfluid than the condensate 
fraction Q). In addition, it is predicted that for any 
value of na 3 there exists a critical strength of disorder 
R c = 16/tt ~ 5.1 for which p s /p < N /N. The results 
of the Bogoliubov model are expected to be valid for di- 
lute systems and weak disorder. However, it is not clear 
whether these results still apply for R > R c in a range 
of densities where the difference between p s /p and No/N 
can be significant. These questions have been addressed 
using the DMC method. 

DMC simulation. We consider a system of N spin- 
less bosons of mass m and iVj mp impurities placed 
at random in a box with periodic boundary con- 
ditions. The Hamiltonian of the system is given 
by H = -(ft 2 /2m)£tiV? + u (| r< - r .|) + 

EfcLi Sfci P v (\ r i ~ r e\)i where u(r) and v(r) are re- 
spectively the particle-particle and particle- impurity two- 
body potential. For both potentials we use a hard- 
sphere model: particles have diameter a and impu- 
rities have diameter 2b — a, where b is the range of 
v(r). Impurities have fixed position rg and overlap be- 
tween impurities is avoided. Importance sampling is used 
through the trial wavefunction ^>y(R) = ^t(i"i, ..,rjv) = 
Hi<j f{ r ij)Hi,e9{ru)- The Jastrow factors, f(r) of a 
pair of particles and g(r) of a particle-impurity pair, are 
calculated using the same technique as in Ref. [12] . Aver- 
age over disorder is obtained by repeating the simulation 
for different configurations of impurities. A number be- 
tween 5 and 10 independent configurations has proven to 
be enough. The direct output of the DMC algorithm is 
the ground-state energy, which is exact apart from statis- 
tical uncertainty (for further details on the DMC method 
see Ref. 0). The superfluid fraction p s /p can be cal- 
culated by extending to zero temperature the winding- 
number technique employed in PIMC calculations p4[ , 



2 



as discussed for bosons on a lattice in Ref. ||. The su- not see this behavior. In fact, although the agreement 
perfluid fraction is obtained as the ratio of two diffusion between p s /p and Eq. (|J) is good up to relatively large 
constants p s /p — D s /Dq, where Do — h 2 /2m is the dif- values of na 3 , the depletion of the condensate becomes 



fusion constant in imaginary time of a free particle and 



D. = 



N J dK /(R,r)[fl CM (r) - Rcm{0)} 2 
™6t /dR/(R,r) 



(6) 



is the diffusion constant of the "center of mass" of the 
system Rqm = 0-/N) J2i=i r «- ^ n tne above equation 
/(R, t) is the probability density of walkers generated 
by the DMC algorithm during integration in imaginary 
time t. One can prove that the above result for p s /p 
is exact and does not depend on the choice of the trial 
wavefunction Jl5| ] . Finally, the condensate fraction is ob- 
tained from the long-range behavior of the one-body den- 
sity matrix: N /N = liirij.-s.oo p( r ) ( see R°f- @ f° r fur- 
ther details). We performed calculations for values of 
N = 16, 32 and 64 and no significative finite-size effects 
were found. 

Results. In Fig. [l], results for the energy beyond mean 
field as a function of the gas parameter and for differ- 
ent strengths of disorder are presented. For R = 2 we 
find good agreement with Eq. (||) over a wide range of 
densities. By increasing R, deviations start to appear 
at lower densities. In particular, for the largest value 
R = 100, we do not find agreement for densities larger 
than na 3 > 10~ 5 . 



10" 



10 1 



10' 



10 1 



10* 




na 



FIG. 1. Energy per particle beyond mean field. The results 
for a given strength of disorder R are obtained for a fixed 
concentration \ and a fixed ratio b/a as shown in the figure. 
The error bars are smaller than the size of the symbols. The 
solid lines correspond to Eq. Q). Energies are in units of 
H 2 /2ma 2 . 

In Fig. ||, we show results for p s J p and Nq/N. For 
R = 2 the supcrfluid fraction follows the analytical pre- 
diction (||) up to large values of na 3 . On the contrary, 
the condensate fraction is more sensitive to the increase 
of density and deviates earlier from the Bogoliubov re- 
sult (H). The value R — 12.5 corresponds to a strength 
of disorder above the critical value (R c = 5.1), where 
the Bogoliubov model predicts p s /p < Nq/N. We do 



very soon larger than predicted by Eq. (|4j) and, as a 
consequence, we find either p s /p ~ Nq/N at very low 
densities or p s /p > Nq/N for larger densities. The re- 
sults for R = 100 correspond to a regime of strong dis- 
order where Bogoliubov model can not be applied. In 
this regime, p s /p and N /N first decrease together with 
increasing density and then, for na 3 > 10~ 4 , a clear gap 
appears with the superfluid fraction significantly smaller 
than the condensate fraction. To our knowledge this is 
the first direct realization of a system exhibiting this un- 
usual feature. 




FIG. 2. Superfluid fraction p s /p (solid symbols) and con- 
densate fraction No/N (open symbols). Disorder parameters 
are as in Fig. Jl| Solid lines correspond to Eq. (||) and dashed 
lines to Eq. (|). 

The cross-over from weak to strong disorder is better 
shown in Fig. [3| In the figure we present results for p s j ' p 
and No/N as a function of R at the density na 3 = 10~ 4 . 
By increasing the strength of disorder, superfluid and 
condensate fractions first decrease together, and for large 
values of R the strong disorder regime of Fig. || where 
Ps/p < No/N is achieved. At large density the situation 
is different as shown in Fig. |J, where na 3 = 10~ 2 . Al- 
ready in the absence of disorder interaction effects give 
rise to a sizable depletion of the condensate (about 20%) 
and by adding disorder no clear evidence of a regime 
where p s /p < No/N is observed. An interesting result 
which emerges from Figs. |[| is that the behavior of the 
superfluid fraction is well described by the Bogoliubov 
prediction (||) also for high densities, provided R is small. 
On the contrary, the condensate fraction is much more 
sensitive to the value of the gas parameter and agreement 
with (0) is found only in the regime where both na 3 < < 1 
and Vna^R << 1. 

In the regime where No/N and p s / p agree with the an- 
alytical predictions [results (Q), the scaling behavior 
on the parameter R is evident. An important result of 
our analysis concerns the fact that the scaling behavior 
extends well beyond the region where results (0) and (0) 



3 



rply. This is explicitly shown in the insets of Fig. || and 
where we vary both the ratio b/a and the concentra- 
tion x with R = x (b/a) 2 fixed. At small density (Fig. ||) 
we find that, even in the case of strong disorder R = 100, 
deviations from scaling are relatively small. At large den- 
sity (Fig. |I]) we still find good scaling for R = 2, whereas 
for R = 4 a dependence on the value of b/a becomes 
evident. 







8 


R = 25 
• 

o 


8 


♦0 


o 
♦ 


R = 100 
o 
♦ 





100 



150 



200 



FIG. 3. Superfluid fraction (solid symbols) and condensate 
fraction (open symbols) for na = 1CT 4 . The strength of 
disorder R has been varied by changing the concentration 
X of impurities with a fixed ratio b/a = 5. The solid line 
corresponds to Eq. (^) and the dashed line to Eq. (^). Inset. 
Scaling behavior as a function of the ratio b/a for given values 
of the strength R. Error bars have approximately the size of 
the symbols. 




FIG. 4. Superfluid fraction (solid symbols) and condensate 
fraction (open symbols) for na = 10~ 2 . The strength of 
disorder R has been varied by changing the concentration 
X of impurities with a fixed ratio b/a — 2. The solid line 
corresponds to Eq. (^) and the dashed line to Eq. (^). Inset. 
Same as in Fig. H. 



Due to the constraint of non-overlapping impurities 
systems with larger strengths of disorder can not be stud- 
ied. Nevertheless, we have investigated the occurrence 



of a quantum phase transition by analyzing the depen- 
dence of the results for p s /p and N /N on the size of 
the system. Our DMC calculations show no significant 
finite-size effects and the results shown in Figs. ||,[| are 
thus appropriate to the thermodynamic limit. We con- 
clude that within our model of non-overlapping impuri- 
ties there is no quantum phase transition for a critical 
value of disorder. 

In conclusion, we have investigated BEC and superflu- 
idity in a Bose gas with disorder as a function of density 
and strength of disorder. We have shown that dilute 
systems with weak disorder can be correctly described 
using the Bogoliubov model. For strong disorder we find 
that the system exhibits the unusual feature of a super- 
fluid fraction significantly smaller than the condensate 
fraction, in qualitative agreement with the prediction of 
Bogoliubov model. 

The authors would like to thank L.P. Pitaevskii and S. 
Stringari for many useful discussions. One of us (S.G.) 
acknowledges the hospitality of the Aspen Center for 
Physics. This research has been partially supported by 
DOES (Spain) Grant No. PB98-0922. We also acknowl- 
edge supercomputer facilities provided by CEPBA. 



[1] M. Ma et al, Phys. Rev. B 34, 3136 (1986); M.P.A. 

Fisher et al, Phys. Rev. B 40, 546 (1989); K.G. Singh 

and D.S. Rokhsar, Phys. Rev. B 49, 9013 (1994). 
[2] Y.-H. Li and S. Teitel, Phys. Rev. B 41 11388 (1990); 

R.T. Scalettar et al, Phys. Rev. Lett. 66, 3144 (1991); 

W. Krauth et al, Phys. Rev. Lett. 67, 2307 (1991); K. 

Moon and S.M. Girvin, Phys. Rev. Lett. 75, 1328 (1995). 
[3] S. Zhang et al, Phys. Rev. Lett. 74, 1500 (1995). 
[4] K. Huang and H.-F. Meng, Phys. Rev. Lett. 69, 644 

(1992). 

[5] S. Giorgini et al, Phys. Rev. B 49, 12938 (1994). 

[6] A.V. Lopati n and V.M. Vinokur, preprint cond 



mat/0109546 



[7] M. Boninsegni and H.R. Clyde, J. Low Temp. Phys. 112, 
251 (1998). 

[8] M.C. Gordillo and D.M. Ceperley, Phys. Rev. Lett. 85, 
4735 (2000). 

[9] For a review see J.D. Reppy, J. Low Temp. Phys. 87, 
205 (1992); see also G.K.S. Wong et al, Phys. Rev. B 
48, 3858 (1993); P.A. Crowell et al, Phys. Rev. B 51, 
12721 (1995); P.A. Crowell et al, Phys. Rev. Lett. 75, 
1106 (1995). 

[10] H.R. Clyde et al, Phys. Rev. Lett. 84, 2646 (2000). 
[11] F. Dalfovo et al, Rev. Mod. Phys. 71, 463 (1999). 
[12] S. Giorgini et al, Phys. Rev. A 60, 5129 (1999). 
[13] J. Boronat and J. Casulleras, Phys. Rev. B 49, 8920 
(1994). 

[14] E.L. Pollock and D.M. Ceperley, Phys. Rev. B 36, 8343 
(1987). 

[15] G.E. Astrakharchik, Laurea Thesis, unpublished. 



4